Development of innovative deep learning approaches in the analysis of digital pathology images

Deep learning, one of the most popular machine learning methods of recent times, has been successfully applied in many areas such as image processing, voice recognition, signal processing. Deep learning methods are used effectively in autonomous aerial vehicles, robotics, smartphones, and medicine. Computer-aided diagnosis has become a trendy research field with the development of special scanners and digitization of the biopsy process. Health data stored in computer environments continues to increase day by day. Due to this increasing amount of data, the performances of machine learning methods have been significantly improved. Deep learning methods, using digitized images, make it easier for experts to diagnose diseases. Also, deep learning is hope for the automatic diagnosis of critical conditions such as cancer. Breast cancer is one of the most common types of cancer in the community. Early diagnosis of breast cancer significantly increases the success rate in the treatment of a patient. Histopathological images play an essential role in the diagnosis of breast cancer. In this thesis, some innovative deep learning methods have been developed to analyze breast cancer using histopathological images. In the thesis study, several open-access cancer datasets were examined. Various programming packages have been created to be used to analyze these datasets, and some novel techniques have been developed to organize the dataset. In order to analyze the histopathological images, the stages of creating different deep learning models are discussed in detail. Various parameters and methods were examined to obtain the best solution with the developed models. The performances of some pre-trained popular deep models in the problem of classification and segmentation of histopathological images were evaluated. In addition to these, a new deep learning model has been presented in order to reduce the size of digital images reaching high dimensions and to obtain many lower-dimensional representations of these images. Thus, it paved the way for helping to easily transfer high-dimensional digital images in the healthcare field to remote units. In addition, thanks to these methods, classification and labeling of histopathological images were provided at an expert level.